Re: Meeting Agenda 2008-09-08

[Tom Tromey <tromey at redhat dot com>]

Tom> 1. Individual status on estimates and work.

At the meeting we talked a bit about linkage names, mangling, and
dwarf.

I asked on irc and Doug reminded me of the odd test case:

    template<int C> int foo(int (*x)[C + 1]) { return 5; }
    int m() { return foo<2> (0); }

Compile this and 'nm' the .o and you will see:

    00000000 W int foo<2>(int (*) [(2) + (1)])


During the meeting I confused a couple things.

I am not sure that this kind of oddity can occur for type names.  So,
perhaps it isn't a problem if we want to make GCC really emit
DW_TAG_template_type_parameter and the like.

Given the existence of cp-name-parser.y, it isn't clear when we would
need this info from GCC.  (Which doesn't make it a bad idea ... even
if gdb doesn't need it, it would be preferable for GCC to follow the
spec.)

I do think that the above test case means getting rid of the linkage
name is hard, at least assuming that the idea is to somehow
reconstruct linkage names from the DWARF.  But probably I'm also
misunderstanding the point of that project?  Or missing something
obvious?  I am not entirely sure how all the parts here fit together,
so feel free to give me a clue :)

Tom